Research Pred Mod Life Batty

7/27/2019 Research Pred Mod Life Batty

1/29

Predictive Modeling for Life Insurance Ways Life Insurers Can Participate in the Business Analytics Revolution

Prepared by

Mike Batty, FSA, CERA

Arun Tripathi, Ph.D.Alice Kroll, FSA, MAAA

Cheng-sheng Peter Wu, ASA, MAAA, FCAS

David Moore, FSA, MAAA

Chris Stehno

Lucas Lau

Jim Guszcza, MAAA, FCAS, Ph.D.

Mitch Katcher, FSA, MAAA

Deloitte Consulting LLPApril 2010


2/29

2

Predictive Modeling for Life InsuranceWays Life Insurers Can Participate in the Business Analytics Revolution

AbstractThe use of advanced data mining techniques to improve decision making has already taken root in

property and casualty insurance as well as in many other industries [1, 2]. However, the application of

such techniques for more objective, consistent and optimal decision making in the life insurance

industry is still in a nascent stage. This article will describe ways data mining and multivariate analytics

techniques can be used to improve decision making processes in such functions as life insurance

underwriting and marketing, resulting in more profitable and efficient operations. Case studies will

illustrate the general processes that can be used to implement predictive modeling in life insurance

underwriting and marketing. These case studies will also demonstrate the segmentation power ofpredictive modeling and resulting business benefits.

Keywords: Predictive Modeling, Data Mining, Analytics, Business Intelligence, Life Insurance Predictive

Modeling


3/29

3


ContentsThe Rise of Analytic Decision Making........................................................................................................... 4

Current State of Life Insurance Predictive Modeling ................................................................................. 6

Business Application that Can Help Deliver a Competitive Advantage........................................... 10

Life Underwriting ............................................................................................................................................... 10

Marketing............................................................................................................................................................... 14

In-force Management ........................................................................................................................................ 15

Additional Predictive Model Applications ............................................................................................... 16

Building a Predictive Model .............................................................................................................................. 17Data .......................................................................................................................................................................... 17

Modeling Process................................................................................................................................................ 19

Monitoring Results ............................................................................................................................................ 24

Legal and Ethical Concerns ................................................................................................................................ 24

The Future of Life Insurance Predictive Modeling .................................................................................. 26


4/29

4


The Rise ofAnalytic Decision MakingPredictive modeling can be defined as the analysis of large data sets to make inferences or identify

meaningful relationships, and the use of these relationships to better predict future events [1,2]. It uses

statistical tools to separate systematic patterns from random noise, and turns this information into

business rules, which should lead to better decision making. In a sense, this is a discipline that actuaries

have practiced for quite a long time. Indeed, one of the oldest examples of statistical analysis guiding

business decisions is the use of mortality tables to price annuities and life insurance policies (which

originated in the work of John Graunt and Edmund Halley in the 17th century). Likewise, throughout

much of the 20th century, general insurance actuaries have either implicitly or explicitly used

Generalized Linear Models [3,4,5] and Empirical Bayes (a.k.a. credibility) techniques [6,7] for the pricing

of short-term insurance policies. Therefore, predictive models are in a sense, old news. Yet in recent

years, the power of statistical analysis for solving business problems and improving business processes

has entered popular consciousness and become a fixture in the business press. Analytics, as the field

has come to be known, now takes on a striking variety of forms in an impressive array of business and

other domains.

Credit scoring is the classic example of predictive modeling in the modern sense of business analytics.

Credit scores were initially developed to more accurately and economically underwrite and determine

interest rates for home loans. Personal auto and home insurers subsequently began using credit scores

to improve their selection and pricing of personal auto and home risks [8,9]. It is worth noting that one

of the more significant analytical innovations in personal property-casualty insurance in recent decades

originated outside the actuarial disciplines. Still more recently, U.S. insurers have widely adopted

scoring models often containing commercial credit information for pricing and underwriting complex

and heterogeneous commercial insurance risks [10].

The use of credit and other scoring models represents a subtle shift in actuarial practice. This shift has

two related aspects. First, credit data is behavioral in nature and, unlike most traditional rating

variables, bears no direct causal relationship to insurance losses. Rather, it most likely serves as a proxy

measure for non-observable, latent variables such as risk-seeking temperament or careful

personality that are not captured by more traditional insurance rating dimensions. From here it is a

natural leap to consider other sources of external information, such as lifestyle, purchasing, household,social network, and environmental data, likely to be useful for making actuarial predictions [11, 24].

Second, the use of credit and other scoring models has served as an early example of a widening domain

for predictive models in insurance. It is certainly natural for actuaries to employ modern analytical and

predictive modeling techniques to arrive at better solutions to traditional actuarial problems such as

estimating mortality, setting loss reserves, and establishing classification ratemaking schemes. But


5/29

5

actuaries and other insurance analytics are increasingly using predictive modeling techniques to improve

business processes that traditionally have been largely in the purview of human experts.

For example, the classification ratemaking paradigm for pricing insurance is of limited applicability for

the pricing of commercial insurance policies. Commercial insurance pricing has traditionally been driven

more by underwriting judgment than by actuarial data analysis. This is because commercial policies arefew in number relative to personal insurance policies, are more heterogeneous, and are described by

fewer straightforward rating dimensions. Here, the scoring model paradigm is especially useful. In

recent years it has become common for scoring models containing a large number of commercial credit

and non-credit variables to ground the underwriting and pricing process more in actuarial analysis of

data, and less in the vagaries of expert judgment. To be sure, expert underwriters remain integral to the

process, but scoring models replace the blunt instrument of table- and judgment-driven credits and

debits with the precision tool of modeled conditional expectations.

Similarly, insurers have begun to turn to predictive models for scientific guidance of expert decisions in

areas such as claims management, fraud detection, premium audit, target marketing, cross-selling, and

agency recruiting and placement. In short, the modern paradigm of predictive modeling has made

possible a broadening, as well as a deepening, of actuarial work.

As in actuarial science, so in the larger worlds of business, education, medicine, sports, and

entertainment. Predictive modeling techniques have been effective in a strikingly diverse array of

applications such as:

Predicting criminal recidivism [12]

Making psychological diagnoses [12]

Helping emergency room physicians more effectively triage patients [13]

Selecting players for professional sports teams [14]

Forecasting the auction price of Bordeaux wine vintages [15]

Estimating the walk-away pain points of gamblers at Las Vegas casinos to guide casino

personnel who intervene with free meal coupons [15]

Forecasting the box office returns of Hollywood movies [16]

A common theme runs through both these and the above insurance applications of predictive modeling.

Namely, in each case predictive models have been effective in domains traditionally thought to be in the

sole purview of human experts. Such findings are often met with surprise and even disbelief.

Psychologists, emergency room physicians, wine critics, baseball scouts, and indeed insurance

underwriters are often and understandably surprised at the seemingly uncanny power of predictivemodels to outperform unaided expert judgment. Nevertheless, substantial academic research,

predating the recent enthusiasm for business analytics by many decades, underpins these findings. Paul

Meehl, the seminal figure in the study of statistical versus clinical prediction, summed up his lifes work

thus [17]:


6/29

6

There is no controversy in social science which shows such a large body of quantitatively

diverse studies coming out so uniformly in the same direction as this one. When you

are pushing over 100 investigations, predicting everything from the outcome of football

games to the diagnosis of liver disease, and when you can hardly come up with half a

dozen studies showing even a weak tendency in favor of the clinician, it is time to draw

a practical conclusion.

Certainly not all applications of predictive modeling have a clinical versus actuarial judgment character

[18]. For example, amazon.com and netflix.com make book and movie recommendations without any

human intervention [25]. Similarly, the elasticity-optimized pricing of personal auto insurance policies

can be completely automated (barring regulatory restrictions) through the use of statistical algorithms.

Applications such as these are clearly in the domain of machine, rather than human, learning. However,

when seeking out ways to improve business processes, it is important to be cognizant of the often

surprising ability of predictive models to improve judgment-driven decision-making.

Current State of Life Insurance Predictive ModelingWhile life insurers are noted among the early users of statistics and data analysis, they are absent from

the above list of businesses where statistical algorithms have been used to improve expert-driven

decisions processes. Still, early applications of predictive modeling in life insurance are beginning to

bear fruit, and we foresee a robust future in the industry [19].

Life insurance buffers society from the full effects of our uncertain mortality. Firms compete with each

other in part based on their ability to replace that uncertainty with (in aggregate) remarkably accurate

estimates of life expectancy. Years of fine-tuning these estimates have resulted in actuarial tables that

mirror aggregate insured population mortality, while underwriting techniques assess the relative risk of

an individual. These methods produce relatively reliable risk selection, and as a result have been

accepted in broadly similar fashion across the industry. Nonetheless, standard life insurance

underwriting techniques are still quite costly and time consuming. A life insurer will typically spend

approximately one month and several hundred dollars underwriting each applicant1.

Many marginal improvements to the underwriting process have taken hold: simplified applications for

smaller face amounts, refinement of underwriting requirements based upon protective value studies,

and streamlined data processing via automated software packages are all examples. However, the

examples in the previous section suggest that property-casualty insurers have gone farther in

developing analytics-based approaches to underwriting that make better use of available information to

yield more accurate, consistent, and efficient decision-making. Based on our experience, life insurance

underwriting is also ripe for this revolution in business intelligence and predictive analytics. Perhaps

1According to the Deloitte 2008 LIONS benchmarking study of 15 life insurers, the median service time to issue a

new policy ranges between 30 and 35 days for policies with face amounts between $100k to $5 million, and the

average cost of requirements (excluding underwriter time) is $130 per applicant.

As used in this document, Deloitte means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please seewww.deloitte.com/us/aboutfor a detailed description of the legal structure of Deloitte LLP and its subsidiaries.
http://www.deloitte.com/us/abouthttp://www.deloitte.com/us/abouthttp://www.deloitte.com/us/about


7/29

7

motivated by the success of analytics in other industries, life insurers are now beginning to explore the

possibilities2.

Despite our enthusiasm, we recognize that life underwriting presents its own unique set of modeling

challenges which have made it a less obvious candidate for predictive analytics. To illustrate these

challenges it is useful to compare auto underwriting, where predictive modeling has achievedremarkable success, with life underwriting, where modeling is a recent entry. Imagine everything an

insurer could learn about a prospective customer: age, type of car, accident history, credit history,

geographic location, personal and family medical history, behavioral risk factors, and so on. A predictive

model provides a mapping of all these factors combine onto the expected cost of insuring the customer.

Producing this map has several prerequisites:

A clearly defined target variable, i.e. what the model is trying to predict

The availability of a suitably rich data set, in which at least some predictive variables correlated

with the target can be identified

A large number of observations upon which to build the model, allowing the abidingrelationships to surface and be separated from random noise

An application by which model results are translated into business actions

While these requirements are satisfied with relative ease in our auto insurance example, life insurers

may struggle with several of them.

Auto Insurer Life Insurer

Target

VariableClaims over six-month contract

Mortality experience over life of product

(10, 20+ years)

Modeling

Data

Underwriting requirements supplemented

by third-party data

Underwriting requirements supplemented

by third-party dataFrequency

of Loss

Approximately 10 percent of drivers make

claims annually

Typically, fewer than 1 first year death per

1,000 new policies issued

Business

ActionUnderwriting Decision Underwriting Decision

Statisticians in either domain can use underwriting requirements, which are selected based upon their

association with insurance risk, supplement them with additional external data sources, and develop

predictive models that will inform their underwriting decisions. However, the target variable and

volume of data required for life insurance models raise practical concerns.

For the auto insurer, the amount of insurance loss over the six-month contract is an obvious candidatefor a models target variable. But because most life insurance is sold through long duration contracts,

the analogous target variable is mortality experience over a period of 10, 20, or often many more years.

Because the contribution of a given risk factor to mortality may change over time, it is insufficient to

analyze mortality experience over a short time horizon. Further, auto insurers can correct underwriting

2As reported in an SOA sponsored 2009 study, Automated Life Underwriting, only 1 percent of North American

life insurers surveyed are currently utilizing predictive modeling in their underwriting process.


8/29

8

mistakes through rate increases in subsequent policy renewals, whereas life insurers must price

appropriately from the outset.

The low frequency of life insurance claims (which is good news in all other respects) also presents a

challenge to modelers seeking to break ground in the industry. Modeling statistically significant

variation in either auto claims or mortality requires a large sample of loss events. But whereasapproximately 10 percent of drivers will make a claim in a given year, providing an ample data set, life

insurers can typically expect less than one death in the first year of every 1,000 policies issued3. Auto

insurers can therefore build robust models using loss data from the most recent years of experience,

while life insurers will most likely find the data afforded by a similar time frame insufficient for modeling

mortality.

The low frequency of death and importance of monitoring mortality experience over time leaves

statisticians looking for life insurance modeling data that spans many (possibly 20) years. Ideally this

would be a minor impediment, but in practice, accessing usable historical data in the life insurance

industry is often a significant challenge. Even today, not all life insurers capture underwriting data in an

electronic, machine-readable format. Many of those that do have such data only implemented the

process in recent years. Even when underwriting data capture has been in place for years, the contents

of the older data (i.e. which requirements were ordered) may be very different from the data gathered

for current applicants.

These challenges do not preclude the possibility of using predictive modeling to produce refined

estimates of mortality. However, in the short term they have motivated a small, but growing number of

insurers to begin working with a closely related yet more immediately feasible modeling target: the

underwriting decision on a newly issued policy. Modeling underwriting decisions rather than mortality

offers the crucial advantage that underwriting decisions provide informative short term feedback in high

volumes. Virtually every application received by a life insurer will have an underwriting decision

rendered within several months. Further, based upon both historical insurer experience and medical

expertise, the underwriting process is designed to gather all cost-effective information available about

an applicants risk and translate it into a best estimate offuture expected mortality. Therefore, using

the underwriting decision as the target variable addresses both key concerns that hinder mortality-

predicting models.

Of course, underwriting decisions are imperfect proxies for future mortality. First, life underwriting is

subject to the idiosyncrasies, inconsistencies, and psychological biases of human decision-making.

Indeed this is a major motivation for bringing predictive models to bear in this domain. But do these

idiosyncrasies and inconsistencies invalidate underwriting decisions as a candidate target variable? No.

To the extent that individual underwriters decisions are independent of one another and are not

affected by common biases, their individual shortcomings tend to diversify away. A famous example

3This is an estimate based upon industry mortality tables. Mortality experience varies across companies with

insured population demographics. In the 2001 CSO table, the first-year select, weighted average mortality rate

(across gender and smoker status) first exceeds 1 death per thousand at age 45.


9/29

9

illustrates this concept. When Francis Galton analyzed 787 guesses of the weight of an ox from a contest

at a county fair, he found that the errors of the individual guesses essentially offset one another, and

their average came within one pound of the true weight of the ox. This illustrates how regression and

other types of predictive models provide a powerful tool for separating signal from noise.

In fact, the Galton example is quite similar to how life insurers manage mortality. Although individual

mortality risk in fact falls along a continuum, insurers group policyholders into discrete underwriting

classes and treat each member as if they are of average risk for that class. When the risks are segmented

sufficiently, insurers are able to adequately price for the aggregate mortality risk of each class.

However, to avoid anti-selection and maintain the integrity of the risk pools insurers must segment risks

into classes that are homogenous. While the noise in underwriting offers may diversify, these offers

are accepted or rejected by applicants strategically. On average, applicants who have knowledge of their

own health statuses will be more likely to accept offers that are in their favor, and reject those that are

disadvantageous. For example, in the figure below an applicant at the upper range of the standard class

may qualify for preferred with another insurer, thus leaving the risk profile of the original standard class

worse than expected.

Therefore, anything that widens the range of mortality risks in each class, and thus blurs the lines

between them, poses a threat to a life insurer. In addition to the inconsistency of human decision

making, global bias resulting from company-wide underwriting guidelines that may not perfectly

represent expected mortality can also contribute to this potential problem.


10/29

10

While modeling underwriting decisions may ultimately become a step along the path towards modeling

mortality directly, we do believe today it is a pragmatic approach that provides the maximal return on

modeling investment today. Specifically, utilizing underwriting decisions as the target variable is

advantageous because they are in generous supply, contain a great deal of information and expert

judgment, and do not require long development periods as do insurance claims. At the same time

they contain diversifiable noise that can be dampened through the use of predictive modeling.

Although building models for mortality and improving risk segmentation remain future objectives,

utilizing predictive models based upon historical underwriting decisions represents a significant

improvement on current practice, and is a practical alternative in the common scenario where mortality

data is not available in sufficient quantities for modeling.

Business Application That Can Help Deliver a Competitive AdvantageWe will describe the technical aspects of underwriting predictive models in some detail in a subsequent

section. While that discussion may beguile certain members of the audience (the authors included),

others will be more interested in understanding how predictive modeling can deliver a competitiveadvantage to life insurers.

Life Underwriting

Unsurprisingly, one compelling application has been to leverage models that predict underwriting

decisions directly within the underwriting process. As mentioned above, underwriting is a very costly

and time consuming, but necessary, exercise for direct life insurance writers. Simply put, the

underwriting process can be made faster, more economical, more efficient, and more consistent when a

predictive model is used to analyze a limited set of underwriting requirements and inexpensive third-

party marketing data sources (both described below) to provide an early glimpse of the likelyunderwriting result. As illustrated in Figure 1, the underwriting predictive models that Deloitte has

helped insurers develop have been able to essentially match the expected mortality for many

applicants. These insurers are beginning to leverage model results to issue many of their policies in just

several days, thus foregoing the more costly, time consuming, and invasive underwriting requirements.


11/29

11

Figure 1: Mortality of Predictive Model vs. Full Underwriting

Risks which had been underwritten by the insurer and kept in a holdout sample were rank-ordered by

model score and divided into equal-sized deciles. Modeled mortality is computed by taking a

weighted average of the insurers mortality estimates for each actual underwriting class in proportion

to their representation within each decile. Pricing mortality represents the fully underwriting pricing

mortality assumptions.

Issuing more policies with fewer requirements may initially seem like a radical change in underwriting

practices, but we think of it as an expansion of a protective value study. Just as insurers currently mustjudge when to begin ordering lab tests, medical exams records, and EKGs, the models are designed to

identify which applicant profiles do and do not justify the cost of these additional requirements. Based

on the results of the models weve helped insurers build thus far, the additional insight they provide has

allowed these insurers to significantly change the bar on when additional tests are likely to reveal latent

risk factors. As indicated by the quality of fit between the model mortality and pricing assumptions,

these models have been able to identify approximately 30 percent to 50 percent of the applicants that

can be issued policies through a streamlined process, and thus avoid the traditional requirements.

With impressive frequency, the underwriting decision recommended by these models matched the

decision produced through full underwriting. For cases when they disagree, however, we offer two

possibilities: 1) the models do not have access to information contained in the more expensive

requirements which may provide reason to change the decision, or 2) models are not subject to biases

or bounded cognition in the same way that underwriters, who do not always act with perfect

consistency or optimally weigh disparate pieces of evidence, are. The latter possibility comports with

Paul Meehls and his colleagues studies of the superiority of statistical over clinical decision making, and

is further motivation for augmenting human decision-making processes with algorithmic support.

X


12/29

12

In our analyses of discrepancies between models and underwriting decisions we did encounter cases

where additional underwriting inputs provided valuable results, but they were rivaled by instances of

underwriting inconsistency. When implementing a model, business rules are used to capitalize upon the

models ability to smooth inconsistency, and channel cases where requirements are likely to be of value

to the traditional underwriting process. Thus, our experience therefore suggests that insurance

underwriting can be added to the Meehl schools long list of domains where decision-making can be

materially improved through the use of models.

These results point to potentially significant cost savings for life insurers. Based on a typical companys

volume, the annual savings from reduced requirements and processing time are in the millions, easily

justifying the cost of model development. Table 1 shows a rough example of the potential annual

savings for a representative life insurer. It lists standard underwriting requirements and roughly typical

costs and frequencies with which they would be ordered in both a traditional and a model-enhanced

underwriting process. It then draws a comparison between the costs of underwriting using both

methods.

Table 1: Illustrative Underwriting Savings from Predictive Model

Requirement

Cost

Requirement Utilization

Traditional

Underwriting

Predictive

Model

Paramedical Exam $55 50% 0%

Oral Fluids Analysis $25 20% 0%

Blood and Urine Analysis $55 70% 0%

MVR Report $6 70% 75%

Attending Physician Statement $100 20% 0%

Medical Exam $120 20% 0%EKG $75 10% 0%

Stress Test $450 1% 0%

Third-Party Data $0.50 0% 100%

Total Cost Per Applicant $130 $5

Savings Per Applicant $125

Annual Applications Received 50,000

Annual Savings (over 30% to 50% of applications) $2 to $3 million

In addition to hard dollars saved, using a predictive model in underwriting can generate opportunities

for meaningful improvements in efficiency and productivity. For example, predictive modeling canshorten and reduce the invasiveness of the underwriting. The time and expense required to underwrite

an application for life insurance and make an offer is an investment in ensuring that risks are engaged at

an appropriate price. However, the effort associated with the underwriting process can be considered a

deterrent to purchasing life insurance. Resources spent while developing a lead, submitting an

application, and underwriting a customer who does not ultimately purchase a policy are wasted from

the perspective of both the producer and home office. The longer that process lasts, and the more tests


13/29

13

the applicant must endure, the more opportunity the applicant has to become frustrated and abort the

purchase entirely, or receive an offer from a competitor. Further, complications with the underwriting

process also provide a disincentive for an independent producer to bring an applicant to a given insurer.

Enhancing underwriting efficiency with a model can potentially help life insurers generate more

applications, and place a higher fraction of those they do receive. In addition, the underwriting staff,

which is becoming an increasingly scarce resource4, will be better able to handle larger volumes as more

routine work is being completed by the model.

We should emphasize that we do not propose predictive models as replacements for underwriters.

Underwriters make indispensible contributions, most notably for applicants where medical tests are

likely to reveal risk factors requiring careful consideration. Ideally, models could be used to identify the

higher risk applicants early in the underwriting process, streamline the experience for more

straightforward risks, and thus free up the underwriters time for analysis of the complex risks. In

addition, underwriters can and should provide insight during the construction, evaluation, and future

refinements of predictive models. This is an oft overlooked but significant point. Particularly in complex

domains such as insurance, superior models result when the analyst works in collaboration with theexperts for whom the models are intended.

How exactly does the process work? The rough sequence is that the insurer receives an application,

then a predictive model score is calculated, then a policy is either offered or sent through traditional

underwriting. In more detail, the predictive model is typically used not to make the underwriting

decisions, but rather to triage applications and suggest whether additional requirements are needed

before making an offer. To that end, the model takes in information from any source that is available in

near-real time for a given applicant. This can include third-party marketing data and more traditional

underwriting data such as the application/tele-interview, MIB, MVR, and electronic prescription

database records. For most insurers, this data can be obtained within two days of receiving theapplication5.

We should point out one key change some insurers must endure. It is essential that producers do not

order traditional requirements at the time an application is taken. If all requirements are ordered

immediately at the application, eliminating them based upon model results is impossible. For some

insurers, this is a major process change for the producer group.

After loading the necessary data for model inputs, the model algorithm runs and produces a score for

the application. From here, several approaches can lead to an underwriting decision. One central issue

insurers may wrestle with is how to use the model output when justifying an adverse action (i.e. not

offering an individual applicant the lowest premium rate). Due to regulatory requirements and industryconventions, it is customary to explain to applicants and producers the specific reasons in cases where

4According to the Bureau of Labor Statistics 2010-2011 Occupational Outlook Handbook, despite reduced

employment due to increased automation, the job outlook of insurance underwriters is classified as good

because the need to replace workers who retire or transfer to another occupation will create many job openings.5

Receiving the application is defined as when all application questions have been answered and/or the tele-

interview has been conducted. If applicable, this includes the medical supplement portion of the application.


14/29

14

the best rates are not offered. It is possible to fashion a reason message algorithm that decomposes

the model score into a set of intuitively meaningful messages that convey the various high-level factors

pushing an individual score in a positive or negative direction. There is considerable latitude in the

details of the reason message algorithm, as well as the wording of the messages themselves.

While allowing the model algorithm to place applicants in lower underwriting classes while deliveringreason codes is a viable, given the novelty of using predictive modeling in underwriting, the approach

life insurers have been most comfortable with thus far is underwriting triage. That is, allowing the model

to judge which cases require further underwriting tests and analysis, and which can be issued

immediately. From a business application perspective, the central model implementation question then

becomes: what model score qualifies an applicant for the best underwriting class that would otherwise

be available based upon existing underwriting guidelines? The information contained in the application

and initial requirements will set a ceiling upon the best class available for that policy. For example, let

us assume an insurer has set an underwriting criterion that says children of parents with heart disease

cannot qualify for super preferred rates. Then for applicants that disclose parents with this condition on

the application, a model can recommend an offer at preferred rates without taking the decisive step inthe disqualification from super preferred.

That is, the role of the model is to determine whether an applicants risk score is similar enough to other

applicants who were offered preferred after full underwriting. If so, the insurer can offer preferred to

this applicant knowing the chance that additional requirements will reveal grounds for a further

downgrade (the protective value) will be too small to justify their cost. If the applicants risk score is not

comparable to other preferred applicants, the insurer can continue with the traditional underwriting.

Marketing

In addition to making the underwriting process more efficient, modeling underwriting decisions can be

of assistance in selling life insurance by identifying potential customers who are more likely to qualify for

life insurance products. Marketing expenses are significant portions of life insurance company budgets,

and utilizing them efficiently is a key operational strategy. For example, a company may have a pool of

potential customers, but know little about their health risks at the individual level. Spreading the

marketing efforts evenly over the pool will yield applicants with average health. However, this company

could likely increase sales by focusing marketing resources on the most qualified customers.

The models supporting underwriting decisions that we have discussed thus far leverage both third-party

marketing data and a limited set of traditional underwriting requirements. Alternatively, we can build

predictive models using only the marketing data. While these marketing models do not deliver thesame predictive power as those that utilize traditional underwriting data, they still segment risks well

enough to inform direct marketing campaigns. Scoring the entire marketing pool and employing a

targeted approach should help reduce the dollars spent marketing to those who will later be declined or

less likely to accept an expensive offer, and result in an applicant pool that contains more healthy lives.


15/29

15

Figure 2: Marketing Model Segmentation

Like Figure 1, risks which had been underwritten by the insurer and kept in a holdout sample were

rank-ordered by model score (using third-party data only) and divided into equal-sized deciles.

However, this graph shows fractions of those deciles which contain declined or substandard

applicants.

In addition to general target marketing efforts, models of underwriting decisions can also serve more

specific sales campaigns. For example, multiline insurers, or even broader financial institutions oftenattempt to increase sales by cross-marketing life products to existing customers. However, they run the

risk of alienating a current customer if the post-underwriting offer is worse than what the marketing

suggests. Instead of selling an additional product, the company may be at risk of losing the customer. In

dealing with this challenge, predictive modeling can be used to conduct an initial review of the customer

pool and assist in determining which existing customers should receive offers for life insurance.

Predictive modeling can also aid in targeting specific products to the markets for which they were

designed. For example, a given company may sell a product with preferred rates that are competitive,

but standard rates that are less attractive. Other products may contain incentives for the insured to

maintain healthy lifestyle. To whom might these products appeal? A person who the model indicates iscurrently living a healthy lifestyle is a prime target for such marketing programs.

In-Force Management

It is well known that the effects of underwriting wear off over time. Lives that were initially healthy may

have degraded, and people who appeared to be poor risks initially may have improved. Products are

priced to anticipate a reversion to mean health risk, but considerable variation in the health status of in-

50%


16/29

16

force policyholders will both remain and be unobservable without new underwriting. While full

underwriting is cost prohibitive in these situations, a predictive model could be an inexpensive and

transparent alternative. Scoring the in-force block could provide more insight to emerging mortality

experience, inform changes to nonguaranteed policy features, help insurers know where to focus efforts

to retain policyholders, and guide both direct writers and reinsurers in block transactions.

Additional Predictive Model Applications

We have focused our discussion on modeling health risk for life insurers because it is arguably the latest

advancement, but there are many other areas of uncertainty for life insurers where a predictive model

could reveal valuable information. We will present several potential applications in brief.

Analogous to models used to market consumer products, predictive algorithms can also estimate how

interested a potential customer would be in purchasing a product from a life insurance company.

Insurance customers are often relatively affluent, or have recently undergone life-changing events such

as getting married, having children, or purchasing a house. All of these traits and events (among other

characteristics) can be identified in the marketing data. More specifically, a predictive model can be

built to identify which characteristics are most highly correlated with the purchase of life insurance.

Again, scoring a direct marketing database can help a life insurer determine where to focus limited

resources for marketing and sales.

We have discussed retention in terms of which customers an insurer would most like to keep, but

equally important is which customers are most likely to leave. As many of the same life event and

lifestyle indicators in the marketing data communicate when a person is likely to purchase a product,

they also hint when a person is likely to surrender a product. In addition to third-party data, insurers

also can see indicators of impending surrenders in transactional data such as how policyholders are

paying premiums (automatic bank debits vs. having to physically write each year, or month), whether a

policyholder is taking policy loans, whether they are calling the home office asking for cash values,

account balances, and in-force illustrations, etc. Since neither producers nor the home office can give

complete attention to each policyholder, a predictive model can sort these different indicators and help

prioritize where to focus policy-saving effort.

Predictive modeling becomes even more powerful when models are used in combination. Not only can

they answer who is most likely to purchase or surrender, but they can simultaneously identify the

customers the company would most like to insure. Layering the underwriting health risk model on top

of either the purchase or surrender models will tell the insurer which quadrant of the population will

likely yield the highest return.


17/29

17

A final application we will mention is workforce analytics. Becoming a successful life insurance agent is

notoriously difficult. The home office spends significant resources recruiting and training these agents,

and the high turnover rate is a considerable drain. Predictive models can be used to help improve the

efficiency of agent recruiting by scoring applicants on the basis of how similar their profile is to that of a

companys existing successful field force. Such a tool can help prioritize which applicants to pursue.

When considering all the potential applications for predictive modeling in life insurance, it becomes

apparent that analytics is truly an enterprise capability rather than a niche solution. After an insurer

begins with one application that proves successful, the next application follows more easily than the

first. Expertise, data, and infrastructure can be leveraged throughout the organization, but more

importantly, decision makers come to realize and respect the power of predictive modeling.

Building a Predictive ModelAfter discussing so much about what can be done with predictive models in life insurance, we have

finally come to how to build one. The following section describes the technical process of developing a

model.

Data

Predictive modeling is essentially an exercise in empirical data analysis. Modelers search through

mountains of data for repeatable, statistically significant relationships with the target (underwriting

decision in this case), and generate the algorithm that produces the best fit. Since it is central to the

modeling process, the best place to begin the technical discussion is with the data.

Data miners prefer to start with a wide lens and filter out potential data sources as necessary. We start

by asking, What data can be obtained for an individual applicant? And then move to questions such as,

Which data elements show a relationship with the target?, Is the penetration of the data enough to

generate statistical significance, Is the correlation strong enough to justify the datas cost?, and

finally, Based upon regulatory and compliance concerns, can the data be included in a predictive model


18/29

18

in the underwriting process? In our experience, working through these questions leads to two different

classes of data: a sub-selection of traditional underwriting requirements, and alternative datasets not

traditionally used in underwriting assessments.

The traditional underwriting requirements incorporated into the predictive models generally meet

several criteria:

Available within the first one to two days after an application is submitted

Transmitted electronically in a machine readable format

Are typically ordered for all medically underwritten applicants

Several of the most common data sources are discussed below. The actual sources used by any

particular life insurer may vary.

Application Data (including part 2 or tele-interview) any piece of data submitted to the company by an

insurance applicant is a candidate for the predictive model. There are two keys to successfully using the

data contained in an insurance application in a model. First, the questions which are easiest to workwith are in a format such as multiple choice, Yes/No, or numerical. However, new text mining

applications are making free form text possible in some situations. Second, the new business process

should capture the application electronically and store the answers in a machine readable format such

as a database. Life insurers who do not have application data in a compatible format face considerable

manual data entry during model build.

MIB When member companies receive an application, they will request a report from the Medical

Information Bureau (MIB). This report includes MIB codes which provide detail on prior insurance

applications submitted to other member companies by the person in question.

MVR The Motor Vehicle Record (MVR) provides a history of driving criticisms, if any, for a given

applicant. This inexpensive and readily available data source provides important information on the

applicants risk profile otherwise unavailable in third-party marketing data. Due to its protective value,

it is also a common underwriting requirement for many life insurers.

Electronic Rx Profile in recent years, several firms have started collecting prescription data records

from pharmacy benefit managers nationwide, compiling by individual, and selling this information to

insurers. Many users are enthusiastic about its protective value, and as a result it is becoming a

standard underwriting requirement for an increasing number of life insurance companies. This is

another interesting source for predictive modeling.

Other traditional underwriting requirements, such as blood and urine analysis, EKGs, medical records

and exam, etc., would add predictive power to a model, but the time and cost to include them may

negate the benefits.

Non-traditional third-party data sets come in a variety of shapes and forms, but most recently we have

seen the application of marketing and consumer credit data from companies such as Equifax and Axiom.

It is important to distinguish between marketing data and the credit score information for which these


19/29

19

consumer reporting agencies are better known. Beyond the credit data, these firms also collect and

distribute consumer information for marketing purposes. Whenever you use your credit card to make a

purchase, or provide your phone number or zip code to the cashier, this data is being collected,

aggregated, and resold.

The third party marketing dataset obtained from the consumer credit company contains thousands offields of data. In contrast to the credit score data, is not subject to the Fair Credit Reporting Act (FCRA)

requirements, and does not require signature authority by the insurance applicant to use it in a model.

For the purposes of constructing a model, the data can be returned without personally identifiable

information. Our experience indicates that using an individuals name and address, the typical match

rate for members of these databases is over 95 percent.

We understand if some people react to this with a feeling of someone looking over your shoulder, and

we discuss some of the ethical concerns of using this data in a later section of this article. Here we will

simply say that while many of these data fields are quite interesting for life underwriting, it is important

to note that model scores are not highly dependent upon any one, or even handful of them. Instead,

the picture painted by this data is viewed holistically, trends are identified that are not necessarily

noticeable to the naked eye, and the overall messages about lifestyle and mortality risk are

communicated. For this reason, it is difficult, if not impossible, to send a powerful message that

misrepresents the applicant, or for the applicant to manipulate the data in a misleading fashion.

Modeling Process

The first step in the model building process is to collect and organize all this data. For several reasons, it

is collected for applications received by the insurer over the past 12 to 18 months. Depending upon the

volume of applications received, this time frame typically produces a sample of underwriting decisions

which will be large enough to sufficiently remove the statistical variation in the model, and ensure the

third-party data available is still relevant. To clarify, the external data technically reflects the applicants

lifestyle today, but is still an accurate representation of them when they applied for insurance provided

that time was in the recent past. Based on our experience, 18 months is about when you may begin to

see material changes in the modeling data, and thus question its applicability to the application date.

The actual collection of the third-party marketing data set for model building is typically a painless

process facilitated by the provider, but the availability of internal historical underwriting data can vary

greatly depending upon individual company practices.

Once the data is collected into one centralized data set and loaded into the statistical package in which

the analysis will be performed, data preparation will provide a solid foundation for model development.Data preparation can be summarized into four steps which are described below:

1) Variable Generation

2) Exploratory Data Analysis

3) Variable Transformation

4) Partitioning Model Set for Model Build


20/29

20

Variable Generation

Variable generation is the process of creating variables from the raw data. Every field of data loaded

into the system, including the target and predictive variables, is assigned a name and a data format. At

times this is a trivial process of mapping one input data field to one variable with a descriptive variable

name. However, this step can require more thought to build the most effective predictive models.

Individual data fields can be combined in ways that communicate more information than the fields doon their own.

These synthetic variables, as they are called, vary greatly in complexity. Simple examples include

combining height and weight to calculate BMI, or home and work address to calculate distance.

However, in our experience some of the most informative predictive variables for life insurance

underwriting are what we call disease-state models. These are essentially embedded predictive models

which quantify the likelihood an individual is afflicted with a particular disease such as diabetes,

cardiovascular, or cancer. The results of these models can then be used as independent predictive

variables in the overall underwriting model. Synthetic variables are where the science and art of

predictive modeling come together. There are well-defined processes which measure the correlationsof predictive variables with a target, but knowing which variables to start from relies more on

experience and intuition.

Exploratory Data Analysis

Before even considering the relationship between independent and dependent variables, it is first

important to become comfortable with the contents of the modeling data by analyzing the distributional

properties of each variable. Descriptive statistics such as min, max, mean, median, mode, and frequency

provide useful insight. This process tells modelers what they have to work with, and informs them of

any data issues they must address before proceeding.

After the initial distributional analysis, the univariate (one variable at a time) review is extended toexamine relationship with the target variable. One-by-one, the correlation between predictive and

target variable is calculated to preview ofeach variables predictive power. The variables that stand out

in this process will be highly correlated with the target, well populated, sufficiently distributed, and thus

are strong candidates to include in the final model.

In addition to paring down the list of potential predictive variables, the univariate analysis serves as a

common sense check on the modeling process. Underwriters, actuaries, and modelers can sit down and

discuss the list of variables which show strong relationships. In our experience, most of the variables

that appear are those which underwriters will confirm are important in their processes. However, some

other variables that are present can be a surprise. In these cases, further investigation into the possibleexplanations for the correlation is advisable.

Variable Transformation

The exploratory data analysis will most likely reveal some imperfections in the data which must be

addressed before constructing the model. Data issues can be mitigated by several variable

transformations:


21/29

21

1) Group excessive categorical values

2) Replace missing values

3) Cap extreme values or outliers

4) Capture trends

To increase the credibility of relationships in the data, it is often helpful to group the values of a given

predictive variable into buckets. For example, few people in the modeling data are likely to have a

salary of exactly $100,000, which means it is difficult to assign statistical significance to the likelihood an

individual with that salary to be underwritten into a particular class. However, if people with salaries

between $90,000 and $110,000 are viewed as a group, it becomes easier to make credible statements

about the pattern of underwriting classes for those people together.

Missing values for different variables among the records in a data set is sometimes problematic.

Unfortunately, there is no simple solution to retrieve the true distribution of variables that have missing

values, but there are several approaches that help mitigate the problem. Modelers could remove all

records in the data set which have missing values for certain variables, but this may be not an ideal

solution because it can create a biased sample or remove useful information. A more common and

effective solution is to replace the missing values with a neutral estimate or a best estimate. The neutral

estimate could be a relatively straightforward metric such as the mean or median value for that variable,

or a more in depth analysis of the best estimate could be the average value for that variable among

other records that most similar to the one in question.

Almost all data sets a modeler encounters in real life will contain errors. A common manifestation of

these errors is extreme values or outliers which distort the distribution of a variable. While not every

outlier is a data error, modelers must weigh the risks and benefits of skewing the overall distribution to

accommodate a very small number of what may or may not be realistic data points. Smoothing these

extreme values may be a poor idea in applications such as risk management where the tail of thedistribution is of utmost concern, but for underwriting predictive modeling it is often worthwhile to

focus more on the center of the distribution. One approach to reducing the distortion is to transform a

variable to a logarithmic scale. While extreme values will be muted, log transformation may minimize

the original trend. Capping extreme values at the highest reasonable value is another simple

alternative.

Finally, transforming variables from text categories to numerical scales can capture trends more readily.

For example, BMI ranges have been officially classified into four categories: under-weight, normal, over-

weight, and obese. Applicants with normal range of BMI are associated with a lower health risk than the

members of the other categories. The trend of the BMI can be captured more effectively bytransforming the BMI categories into an ordinal rank with higher numbers representing higher health

risks, for example, 1=normal, 2=over-weight, 3=under-weight, and 4=obese.

Partitioning Model Set for Model Build

After collecting the data, preparing each variable, and casting aside those variables which will not be

helpful in the model, the data set is divided into three approximately equal parts. Two of these,


22/29

22

commonly called the train and validation sets, are for model building, while the test is placed

aside until the end of the process where it will be used to assess the results [20].

After the data sets are partitioned, modelers carry out an iterative process that produces the strongest

model. Most model builds will test a variety of statistical techniques, but often one effective, and

therefore very common approach, is stepwise regression [21]. This is a fairly complicated process, but inessence, a best fit line that maps a set of predictive variables to the target is created. In a linear model,

this best fit line will be of the form A * variable1 + B * variable2 + = target variable. Variables are

added and removed one-by-one, each time calculating the new best fit line, and comparing the fit of the

new line with the fits of those created previously. This process reveals the marginal predictive power of

each variable, and produces an equation with the most predictive power that relies upon the smallest

number of predictive variables.

Each variable that survives the univariate review should be correlated with the target, but because it

may also be correlated with other predictive variables, not every variable that appears strong on its own

will add marginal value to the model. Among a group of highly correlated variables, stepwise regression

will typically only keep the one or two with the strongest relationships to the target. Another approach

for dealing with highly correlated variables is to conduct a principal components analysis. Similar to the

disease-state models described above, a principal component is a type of sub-predictive model that

identifies the combination of correlated variables which exhibits the strongest relationship with the

target. For example, a principal components analysis of a group of financial variables may reveal that A

* income + B * net worth + C * mortgage principal, and so forth, is a better predictor of underwriting

decision than these variables are on their own. Then result of this equation will then be the input

variable used in the stepwise regression.

The model is first built using the training data, but modelers are also concerned about fitting the model

too closely to the idiosyncratic features of one sample of data. The initial model is adjusted using the

validation data in order to make it more general. Each set is only used once in the modeling process. It

cannot be recycled since the information has already become part of the model; and reusing it would

result in over-fitting.

To assure the model does not reflect patterns in the modeling data which are not repeated in the hold-

out sample, and most importantly, are less likely to be repeated in the applications the company will

receive in the future, the test set is used only to assess the results when modeling is completed. This

step protects predictive modeling from pitfalls like back-testing investment strategies. It is almost

always possible to find a pattern in data looking backwards, but the key question is whether that pattern

will continue in the future. Due to the relative efficiency of financial markets, investment strategieswhich looked so promising in the past usually evaporate in the future. However, in predictive modeling

we generally find that the models built on the train and validation data set hold up quite well for the

test data. The results shown in Figures 1 and 2 are representative of model fit on past test data sets.

At the end of this process modelers will have indentified the equation of predictive variables that has

the strongest statistical relationship with the target variable. A high score from this model implies the


23/29

23

applicant is a good risk, and low score means the opposite. However, this is not the last step in model

development. Layering key guidelines from the existing underwriting process on top of the algorithm is

also a powerful tool. For example, certain serious but rare medical impairments may not occur in the

data with the sufficient frequency to be included in a statistical model, but should not be overlooked by

one either. For these conditions, it can be helpful to program specific rules that a life insurer uses to

govern their underwriting. In addition to acting as a fail safe for rare medical conditions, the

underwriting guidelines can also serve as the basis for making underwriting decisions. In the applications

we have discussed thus far, the model has the authority to determine whether further underwriting is

needed, but not to lower an insurance offer from the best underwriting class. Even for applicants where

the model would recommend a lower underwriting class, incorporating the underwriting guidelines

provides an easily justifiable reason for offering that class.

A final tool to extract useful information out of the modeling data is a decision tree [22]. A decision tree

is a structure that divides a large heterogeneous data set into a series of small homogenous subsets by

applying rules. Each group father along the branches of the tree will be more homogeneous than the

one immediately preceding it. The purpose of the decision tree analysis is to determine a set of if-thenlogical conditions that improve underwriting classification. As a simple example, the process starts with

all applicants, and then splits them based upon whether their BMIs are greater or lower than 30.

Presumably, applicant with BMIs lower than 30 would have been underwritten into a better class than

those with higher BMIs. The stronger variables in the regression equation are good candidates for

decision tree rules, but any of the data elements generated thus far, including predictive variables, the

algorithm score itself, and programmed underwriting rules, can be used to segment the population in

this manner. Figure 3 displays this logic graphically.

Figure 3: Graphical Representation of Decision Tree

In principal, decision trees could be constructed manually, but in practice, dedicated software packages

are much more efficient in identifying the data elements and values upon which to segment the

population. These packages essentially take the brute force approach of trial and error, but due to

computational efficiency they are able to develop optimal multi-node decision trees in manageable

time.


24/29

24

Monitoring Results

In a previous section we discussed how to use the information revealed by predictive models to

generate significant operational efficiencies in the underwriting process. From a technical standpoint,

implementing a predictive modeling application can occur in many different ways. Given the depth of

the topic, this paper leaves these aspects of implementation for a future discussion. However, we

would like to address one area which we believe should be strongly considered a focus after

implementation.

As with traditional underwriting practices, it is critical to monitor the results of a process change. Since

a predictive model is built from a static sample of policyholders who were actually underwritten using

the traditional process, it is important to consider how using it to assess the health risk of a dynamic

population of new applicants may result in anti-selection. Is there potential for applicants and

producers to game the system and exploit the reduced requirements? There are several avenues

through which life insurers can guard against anti-selection.

First, the third party marketing data cannot be easily manipulated by the applicant. It is reporteddirectly by the third-party agency, and is based upon trends captured over time rather than sudden

changes in behavior. Moreover, the model does not rely on any one field from this data, but rather uses

it to form a general understanding about a persons lifestyle. It would be very difficult for an applicant

to systematically alter behavior over time so it presents a false impression. In fact, if the applicant were

successful in systematically altering behavior to change his or her profile, more than likely the

applicants mortality risk would have also changed in the same direction.

To supplement the protection offered by the third party data, it is advisable to maintain a certain degree

of unpredictability in which applicants will be allowed to forgo additional requirements. The

combination of risk factors that qualify an applicant for reduced requirements at each underwriting class

is typically sufficiently complex to offer an initial defense against producers seeking to game the system.

While the patterns are not simple enough to be picked up upon easily, we also recommend a

percentage of applicants who do qualify be selected at random for traditional underwriting. This will

both further disguise the profile of applicants who are eligible for streamlined underwriting, and offer a

baseline for monitoring results. If evidence of anti-selection is present in these applicants, the insurer

will be alerted of the need to alter the process. As in traditional underwriting, producers will seek to

exploit differences in criteria to obtain the best offer for their clients, but this application of predictive

modeling does offer important safeguards against potentially damaging behavior.

Legal and Ethical ConcernsPredictive modeling in life insurance may raise ethical and legal questions. Given the regulations and

norms that govern the industry, these questions are understandable. The authors of this paper are not

legal experts, but we can offer our insight into several issues, and say that in our experience, it is feasible

to assuage these concerns.


25/29

25

Collecting any data about individuals is a sensitive subject. Data collection agencies have been around

since the late 19th century, but in the 1960s lawmakers became concerned with the availability of this

data as they worried that the rapidly developing computing industry would vastly expand its influence,

and lead to potential abuses. This concern resulted in the Fair Credit Reporting Act (FCRA) of 1970. The

essence of the law is that provided certain consumer protections are maintained around access and

transparency, the efficiency gains of making this data available are worthwhile. We tell this story as a

kind of aside because it is the first question asked by many with whom we have discussed predictive

modeling. However, as described above, the data provided by the aggregators come from their

marketing sets which are not subject to the FCRA.

Even though the third-party marketing data does not face explicit FCRA or signature authority legal

restrictions, it can still raise ethical question about whether utilizing the consumer data is overly

invasive. The first point to realize is that commercial use of this personal data is not new. For many

years it has been a valuable tool in selling consumer goods. Marketing firms build personal profiles on

individuals which determine what type of catalogs and mailing advertisements they receive. Google

scans the text of searches and emails in order to present people with related advertisements. Webelieve society has accepted this openness, not without hesitation, because on average it provides more

of what we want, less of what we do not. In addition to consumer marketing applications, predictive

modeling using third-party consumer data has also been accepted for property and casualty insurance

underwriting.

Despite its acceptance in other fields, life insurance has a unique culture, set of norms, and regulations,

so additional care must be taken to use this data in ways that are acceptable. A critical step in predictive

model development is determining which variables to include in the model. We have described the

statistical basis on which these decisions are made, but the process also considers regulatory and

business concerns. Before beginning the model build, the legal and compliance functions of the lifeinsurer should be the first to review the list of potential variables. No matter what their predictive

powers may be, any variable that is deemed to create a legal or public relations risk, or is counter to the

companys values should be excluded from the model. Even if not explicitly forbidden by regulations,

life insurers should err on the side of caution and exclude variables which convey questionable

information, and can feel confident that this caution will not cripple the strength of the model.

The legal and ethical concerns raised also depend upon business decisions that the model is allowed to

influence. While in principal, predictive models could play the lead role in assigning underwriting classes

for many applicants, insurers have been most comfortable from a compliance perspective utilizing

models to triage applications. By using the model as described above to inform the insurer when no

further requirements are needed, the model does not take adverse actions for any applicant. In fact,

the model only has the potential to take a positive action by offering a streamlined underwriting process

that would otherwise be unavailable.

We fully expect and understand that questions will be raised when changes occur to a consumer-facing

process like underwriting. We also recognize that predictive modeling is a new and growing trend in life

insurance, and the industry culture and regulations may evolve to in ways that impact how data and


26/29

26

models are used. For both of these reasons, company legal and compliance experts are key members of

every predictive modeling project we agree to support. While we do not claim to be the definitive

source on this subject, in our experience thus far, it has been possible to utilize predictive modeling for

life insurance underwriting in ways that are compatible with regulatory, ethical, and cultural concerns.

The Future of Life Insurance Predictive ModelingDue to rapid improvements in computation power, data storage capacity, and statistical modeling

techniques, over the last several decades predictive modeling has come into widespread use by

corporations looking to gain a competitive advantage. Banking and credit card industries are well

known pioneers for modeling credit card fraud, personal financial credit score for mortgage and loan

application, credit card mail solicitation, customer cross-sale, and more.

While insurance has lagged behind other industries, more recently it has gained momentum in data

mining and predictive modeling. Early developments include the use of personal financial credit history

for pricing and underwriting for personal automobile and homeowners insurance. As it provedsuccessful in personal lines, predictive modeling has spread into commercial insurance pricing and

underwriting, as well as into a variety of other applications including price optimization models, life-time

customer models, claim models, agency recruiting models, and customer retention models. In just the

last several years, predictive modeling is beginning to show promise in the life insurance industry.

Until relatively recently, merely using predictive models to support underwriting, pricing and marketing

gave property and casualty insurance companies a competitive edge. However, data analytics has

sufficiently penetrated the market so first mover advantages no longer exist. Property and casualty

companies must now improve their modeling techniques and broaden the applications to stay ahead of

their competition [23]. Because application of data mining and predictive modeling is, for the most part,

still new and unexplored territory in life insurance, we do believe those who act first will realize similar

first mover gains.

Our experience indicates that using predictive modeling for underwriting can empower life companies

to segment and underwrite risks through a more consistent and less expensive process. In doing so,

they can reduce costs, improve customer and producer experience, and generate substantial business

growth. Tomorrow, we anticipate those who ignore this emerging trend will scramble to catch up while

the initial users have moved to models of mortality. As a first step in modeling mortality directly, we

have experimented with modeling the main cause of death in the short-term, accidents. At younger

ages, insured mortality is driven by accidental death rather than by disease6. A sample model we have

built to segment which members of a population have been involved in severe auto accidents has shown

substantial promise, and is being incorporated into the latest projects we have supported. The more we

6According to the National Center for Health Statistics National Vital Statistics Reports from March 2005, the top

three causes of death among young adults aged 25-29 are each acute injuries. These account for 61.58 percent of

all deaths at those ages. The leading cause of death is accidental injury (34.09 percent), followed by homicide (14

percent), and suicide (13.49 percent).


27/29

27

discuss full-scale models of mortality with insurers, the more excited they become about their potential,

and committed to unearthing the data to make them a reality. We believe that day is near.

We would like to close by noting that improvements to efficiency and risk selection will not only accrue

to insurers, but also to individuals. Over time, competition will drive insurers to not only capture

additional profits from their reduced costs, but also charge lower premiums and require fewer medicaltests. Because the predictive models we describe do not disadvantage individual applicants, we believe

the long run effect of predictive modeling will be to increase access to insurance. And if the final effect

of predictive modeling in life underwriting is in some small way to push people toward healthier

lifestyles, we would be happy to claim that as the ultimate victory.

This publication contains general information only and is based on the experiences and research of Deloitte practitioners. Deloitte is

not, by means of this publication, rendering business, financial, investment, or other professional advice or services. This publication

is not a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that may affect

your business. Before making any decision or taking any action that may affect your business, you should consult a qualified

professional advisor. Deloitte, its affiliates, and related entities shall not be responsible for any loss sustained by any person whorelies on this publication.

Copyright 2010 Deloitte Development LLC, All rights reserved.


28/29

28

References

1. Davenport, T. H., Harris, J. G., Competing on Analytics: The New Science of Winning,

Harvard Business School Press, (2006).

2. Guszcza, J., Analyzing Analytics, Contingencies, American Academy of Actuaries, July-August, (2008).

3. McCullagh, P., Nelder, J. A., Generalized Liner Models, 2nd Edition, Chapman and Hall,

(1989).

4. Brockman, M. J., Wright, T. S., Statistical Motor Rating: Making Effective Use of Your

Data,Journal of the Institute of Actuaries, Vol. 119, Part III, (1992).

5. Mildenhall, S.J., A Systematic Relationship between Minimum Bias and Generalized

Linear Models, Proceedings of Casualty Actuarial Society, Vol. LXXXVI, CasualtyActuarial Society, (1999).

6. Bailey, R. A., Simon, L. J., An Actuarial Note on the Credibility of a Single Private

Passenger Car, Proceedings of Casualty Actuarial Society, Vol. LVII, Casualty Actuarial

Society, (1959).

7. Bulhmann, H., Experience Rating and Credibility,ASTIN Bulletin IV, Part III, (1967).

8. Insurance Scoring in Private Passenger Automobile Insurance Breaking the Silence,

Conning Report, Conning, (2001).

9. Report on the Use of Credit History for Personal Line of Insurance, American Academy of

Actuaries, (2002).

10.The Top 10 Casualty Actuarial Stories of 2009,Actuarial Review, Vol. 37, No. 1,

Casualty Actuarial Society, (2010).

11.Guszcza, J., Wu, C. P., Mining the Most of Credit and Non-Credit Data, Contingencies,

American Academy of Actuaries, March-April, (2003).

12.Meehl, P., Clinical Versus Statistical Prediction: A Theoretical Analysis and a Review of

the Evidence, University of Minnesota Press, (1954).

13.Gladwell, M., Blink: The Power of Thinking without Thinking, Little Brown and Co, (2005).


29/29

14.Lewis, M., Moneyball: the Art of Winning an Unfair Game, W. W. Norton & Company,

(2003).

15.Ayres, I., Super Crunchers: Why Thinking-by-Numbers Is the New Way to Be Smart,

Bantam Books, (2007).

16.Gladwell, M., The Formula, The New Yorker, (2006).

17.Meehl P., Causes and Effects of My Disturbing Little Book,Journal of Personality

Assessment, Vol. 50, (1986).

18.Dawes, R., Faust, D., Meehl, P., Clinical Versus Actuarial Judgment, Science, 243:

1668-1674, (1989).

19.Guszcza, J., Batty, M., Kroll, A., Stehno, C., Bring Predictive Models to Life, Actuarial

Software Now - A Supplement to Contingencies, American Academy of Actuaries,Winter, (2009).

20.Hastie, T., Tibshirani, R., Friedman, J. H., The Elements of Statistical Learning, Springer,

(2003).

21.Berry, M. J. A., Linoff, G. S., Data Mining Techniques: For Marketing, Sales, and Customer

Relationship Management, John Wiley & Sons, (2003).

22.Breiman, L., Friedman, J., Olshen, R., Stone, C., Classification and Regression Trees, New

York: Chapman and Hall, (1993).

23.Yan, J., Masud, M., Wu, C. P., Staying Ahead of the Analytical Competitive Curve:

Integrating the Broad Range Applications of Predictive Modeling in a Competitive

Market Environment, CAS E-Forum, Winter, Casualty Actuarial Society, (2008).

24.Stehno, C., Johns, C., You Are What You Eat: Using Consumer Data to Predict Health

Risk, Contingencies, American Academy of Actuaries, Winter, (2006).

25.Yi Zhang , Jonathan Koren, Efficient bayesian hierarchical user modeling for recommendationsystem, Proceedings of the 30th annual international ACM SIGIR conference on Research and

development in information retrieval, July 2007, Amsterdam, The Netherlands

Documents

Research Pred Mod Life Batty